Indexation de l’image vidéo couleur non compressée par le contenu à l’aide de la structure d’arbre R

Abstract

From the beginning of nineties, the indexing and research by content became a very active pole of research ; many commercial and academic systems were proposed to search images. Nowadays, with the evolution of the multimedia, videos database appeared, it is also about indexing and searching for videos by content in the aim to answer to several sectors needs as the remote surveillance, the television and teleconference.

In this paper we propose our approach of videos indexing by the content which abord three main points :

• First off all, we propose to apply a multi-dimensional indexing structure used for the similar images databases that are the R-tree to index videos proceeding in cutting-out into individual images and its temporal segmentation which consists in defining plans then to extract key frames of plans, and finally to construct the summary video ; it corresponds to off-line phase of our system. The user have the possibility to introduce to the system a image or a video request, and on the basis of calculation of similarity distance the system return of results ; it corresponds to on-line phase of the system. • In a second time we propose to apply a structure of generic R-tree on the sequence video in order to minimize its space of storage. • In a third time we should pass to video decoding to answer the requests of users.

Finally, in order to estimate the quality of our approach, we tested our system on a set of videos; the results are commented thus.